基于GMM-PSO混合算法的电磁暂态模型参数校正方法

Parameter Correction for Electromagnetic Transient Simulation Model Based on GMM-PSO Hybrid Algorithm

提出了一种基于高斯混合模型(Gaussian mixture model,GMM)–粒子群优化(particle swarm optimization,PSO)混合算法的电力系统电磁暂态模型参数校正方法,旨在根据实测波形校准仿真模型参数,使模型产生的波形曲线最符合客观现实。参数校正问题可以建模为优化问题,即在符合约束条件的各个参数组合中找到估计误差最小的组合。首先,不同于广泛使用的最小二乘法、卡尔曼滤波等模型驱动的参数校正方法,基于贝叶斯分析中参数随机变量化的思想,通过随机指定关键参数并仿真生成大量曲线,将仿真曲线与实测曲线之间的误差定义为特征量。然后,以GMM为工具,将特征量与关键参数的关系建模为联合概率分布。接着基于GMM的条件概率不变性建立特征量到参数的反向映射,从而得到参数的后验分布,并通过其统计量确定对应特定特征量的参数取值。考虑到校正目标的模糊性,采用粒子群算法对特征量进行闭环校验。最后,使用CloudPSS平台搭建测试算例,验证了算法在高维参数校正问题中的有效性。

This paper proposed an electromagnetic transient model parameter calibration method based on GMM-PSO hybrid algorithm. The purpose was to calibrate the parameters of the simulation model based on the measured waveforms and make the behavior curve produced by the model most in line with objective reality. The parameter calibration problem was modeled as an optimization problem. Specifically, found the smallest estimation error among the parameter combinations that met the constraint conditions. Firstly, different from the widely used ordinary least squares, Kalman filter and other model-driven parameter calibration methods, this paper used the idea of parameter random variability in Bayesian analysis. Then, by generating a large amount of simulation data and using GMM as a tool, the relationship between feature quantities and key parameters was modeled as a joint probability distribution. After that, based on the conditional probability invariance of GMM, the mapping of the feature quantity to the parameter was established to obtain the posterior distribution of the parameter, and the accurate value of the parameter was determined by its statistics. Considering the ambiguity of the correction target, a Particle Swarm Optimization was used to perform a closed-loop check on the feature quantity. Finally, a test case was built using the Cloud PSS to verify the effectiveness of the algorithm in the correction of high-dimensional parameters.